UK 2026: What the Supreme Court Ruling on AI Inventions Means for Life‑science Patents in the United Kingdom

Last reviewed: June 22, 2026

The question of AI patentability UK practitioners have debated for years received its most authoritative answer on 11 February 2026, when the UK Supreme Court handed down a landmark judgment that fundamentally realigned the domestic approach to patenting computer‑implemented inventions with established European Patent Office (EPO) practice. Within weeks, the UK Intellectual Property Office (UKIPO) published updated examination guidelines specifically addressing artificial intelligence inventions. For life‑science companies using AI‑assisted drug discovery, from generative protein design to virtual compound screening, these developments create both a broader pathway to patent protection and a more exacting set of evidentiary demands.

This guide unpacks the legal changes, maps them onto real‑world biotech workflows, and provides practical claim‑drafting templates, prosecution strategy and a freedom‑to‑operate checklist designed for in‑house IP counsel and patent attorneys advising R&D teams.

1. Background and Timeline: From the High Court to the 2026 UKSC Judgment and UKIPO AI Guidelines

Key Dates at a Glance

Why the Prior Test Changed

Under the UK Patents Act 1977, computer programs “as such” are excluded from patentability. For over a decade, UK examiners and courts applied a multi‑step “signpost” test that required applicants to show, at a preliminary stage, that the claimed subject‑matter made a “technical contribution” beyond a mere excluded category. In practice, this test was more restrictive than the EPO’s approach, where the assessment of technical effect is integrated into the inventive step analysis rather than applied as an up‑front gateway filter.

The 2026 UKSC judgment recognised this divergence and concluded that the domestic signpost approach was not mandated by the statute and had led to inconsistent outcomes. By aligning the UK with the EPO’s framework, the Supreme Court removed the preliminary exclusion hurdle. Industry observers expect this shift to be particularly significant for AI inventions in the life sciences, where the technical effect, such as a newly identified drug candidate or an improved biological assay, is often embedded in the output rather than the algorithm itself.

2. The Legal Test Now: What the UKSC Actually Decided on AI Patentability UK

Inventorship Remains a Human Requirement

Can AI be an inventor in the UK? No. The Supreme Court’s earlier 2024 decision in Thaler confirmed that the Patents Act requires a natural person to be named as inventor. The 2026 judgment did not revisit that holding. Where an AI system has been instrumental in generating an invention, the human who devised, directed or verified the AI’s output must be identified as the inventor. This requirement applies equally at the UKIPO and at the EPO.

Patentability Approach: No Automatic Exclusion

The central practical effect of the 2026 ruling is that AI inventions will no longer be refused at a preliminary threshold on the ground that they relate to a computer program “as such.” Instead, examiners must assess the claimed invention using the standard patentability criteria:

• Novelty. The invention must be new over the prior art.
• Inventive step. It must not be obvious to a person skilled in the art, and the assessment of what constitutes the inventive step should integrate the technical contribution, mirroring the EPO’s problem‑and‑solution approach.
• Industrial applicability. The invention must be capable of industrial application.
• Technical contribution. The claims must, taken as a whole, contribute something of a technical character to the art. A purely abstract algorithm or mathematical method that produces no technical effect remains excluded.

Worked Example: AI‑Generated Peptide Sequence

Consider a generative model trained on protein‑ligand binding data that outputs a novel peptide sequence with predicted high affinity for a validated oncology target. Under the pre‑2026 test, a UK examiner might refuse the application at the threshold stage on the basis that the claim relates to a computer program executing an algorithm. Post‑2026, the examiner instead assesses the technical contribution of the claimed peptide, its binding characteristics, therapeutic utility and the enabling disclosure, alongside novelty and inventive step. If the peptide is novel, non‑obvious and adequately disclosed, AI patentability UK standards are met regardless of how the candidate was identified.

3. UKIPO Examination Practice After the Guideline Updates

Core AI Versus Applied AI: Key UKIPO Scenarios

The updated UKIPO AI guidelines distinguish between two broad categories. “Core AI” inventions relate to improvements in the AI technique itself, a new neural network architecture, a more efficient training algorithm, or a novel data‑compression method for model inference. “Applied AI” inventions use an AI technique to solve a problem in a specific technical field, such as drug discovery, medical imaging or materials science.

According to the UKIPO guidelines, patents are available for AI inventions in all fields of technology, provided a technical contribution is present. For core AI, the applicant must demonstrate that the algorithmic improvement produces a concrete technical effect, for example, reduced computational resource consumption or improved accuracy in a defined technical task. For applied AI, the technical effect typically resides in the real‑world outcome: a new compound, a more sensitive diagnostic assay, or an optimised manufacturing process.

Examiner Expectations and Evidence

Merely reciting that an invention “uses AI” or “employs a neural network” is insufficient. Examiners will probe for evidence linking the AI component to a specific technical advantage. In the life‑sciences context, this means applicants should include:

• Comparative experimental data showing the AI‑derived output outperforms prior‑art alternatives (e.g., binding affinity data, selectivity panels).
• Sufficient disclosure of the model, training data characteristics, architecture choices, validation methodology, to enable a skilled person to reproduce the result.
• Claims that anchor the AI output to the technical field rather than claiming the algorithm in the abstract.

Early indications suggest that UKIPO examiners are applying these expectations rigorously, requesting additional experimental evidence during examination where the link between the AI step and the technical effect is not clearly drawn in the specification.

4. Specific Implications for AI‑Assisted Drug Discovery Patents

Patentable Subject‑Matter Versus Excluded Matter

The 2026 changes do not eliminate all exclusion concerns. A pure diagnostic algorithm that processes patient data to output a probability score, without any further technical step, may still be challenged as a method of diagnosis practised on the human body or as a mathematical method “as such.” By contrast, a novel compound identified through AI‑driven virtual screening, or a new use of a known compound discovered through AI target identification, falls squarely within patentable subject‑matter if the standard criteria are met.

Data‑Processing Disclaimers and Sufficiency

Life‑science applicants relying on AI‑generated data face heightened sufficiency scrutiny. If the patent specification does not adequately describe the training dataset, the model architecture, and the validation process, an examiner, or a post‑grant challenger, may argue the disclosure is insufficient for a skilled person to reproduce the invention. Industry observers expect sufficiency objections to become a primary battleground in opposition proceedings involving AI‑assisted drug discovery patents.

Biological Sequence and Composition Claims: Enablement Risks

Where AI generates a library of candidate sequences (antibodies, peptides, or small molecules), the applicant must demonstrate that the claimed scope is enabled. Claiming an entire genus of AI‑predicted sequences without supporting data for representative members risks an enablement objection. The practical approach is to claim specific, validated compounds or narrowly defined families, supported by in vitro or in silico evidence of activity.

EPO vs UK Comparison: Post‑2026 Landscape

Worked Examples

Example 1, De novo protein design. A biotech company uses a diffusion model to design a novel single‑domain antibody with nanomolar affinity for a validated immuno‑oncology target. The claim covers the antibody defined by its amino acid sequence, a pharmaceutical composition comprising that antibody, and a method of treating a specified cancer. Enablement is supported by binding data (SPR), cell‑based activity assays and an in vivo efficacy study. This is patentable subject‑matter under both the EPO and post‑2026 UK frameworks.

Example 2, Virtual screening and lead selection. An AI platform screens a virtual library of 10 million small molecules against a crystal structure of a novel kinase target, identifying 50 lead compounds. The applicant claims five specific compounds with validated activity, a method of treatment, and a computer‑implemented method for the screening process itself that produces a ranked list of candidates with predicted binding free energies. The compound and treatment claims are straightforwardly patentable. The computer‑implemented method claim must demonstrate a technical effect beyond mere data processing, here, the prediction of binding free energy linked to an experimentally validated therapeutic outcome would suffice.

5. Practical Patent Drafting and Prosecution Strategy for AI Patentability UK

Claim Drafting Templates

The following templates illustrate claim structures designed to satisfy both UKIPO and EPO examination standards for AI‑assisted life‑science inventions. These are intended as starting points for legal review and adaptation to specific facts.

Template 1, Method Claim (AI‑assisted compound identification):

“A method of identifying a candidate compound for treating [disease X], comprising: (a) providing a trained machine‑learning model configured to predict binding affinity between candidate molecules and [target protein Y]; (b) inputting a library of candidate molecular structures into the model; (c) selecting one or more candidate compounds having a predicted binding affinity below [threshold value]; and (d) validating the selected candidate compound(s) by [in vitro assay / in vivo model], thereby identifying a candidate compound for treating [disease X].”

Template 2, System Claim:

“A computer‑implemented system for predicting protein–ligand interactions, comprising: a processor; a memory storing a trained neural network model comprising [architecture description]; an input module configured to receive molecular structure data; and an output module configured to generate a ranked list of candidate molecules with predicted binding free energies; wherein the system is configured to [specific technical function, e.g., filter candidates based on ADMET property thresholds].”

Template 3, Compound / Composition Claim:

“A compound of formula [structural formula], or a pharmaceutically acceptable salt thereof, for use in the treatment of [disease X], wherein the compound exhibits [measured property, e.g., IC50 of less than 100 nM against target Y].”

Filing Strategy: UKIPO, EPO or Both?

For most life‑science applicants with international commercial interests, the recommended approach remains filing a PCT application to preserve optionality, entering the European regional phase at the EPO and the UK national phase as needed. The post‑2026 alignment between the UK and EPO frameworks means that a single claim set can now be drafted to satisfy both jurisdictions, reducing the need for separate UK‑specific claims or arguments. Applicants seeking early UK protection (for example, to support an NHS market entry or a UK‑based licensing deal) may consider a direct UK filing or an accelerated examination request under the UKIPO’s Green Channel or Patent Prosecution Highway programmes.

Prosecution Tactics and Appeal Considerations

When an examiner raises a technical‑contribution objection, practitioners should focus arguments on the real‑world outcome enabled by the claimed invention rather than the AI technique itself. Presenting comparative data, showing the AI‑derived compound or process outperforms conventional alternatives, is the most effective prosecution tool. At the EPO, this strategy is well‑established through Boards of Appeal case law; the likely practical effect in the UK is that UKIPO hearing officers will now apply similar reasoning.

Oppositions and Post‑Grant Risk

Opponents may attack AI‑assisted patents on sufficiency grounds, arguing that the specification does not enable a skilled person to reproduce the AI model’s output without undue experimentation. To mitigate this risk, applicants should deposit training datasets and model parameters where feasible, provide detailed descriptions of the training methodology, and include sufficient experimental examples to demonstrate that the claimed invention works as described.

6. Freedom‑to‑Operate and Compliance Checklist for AI in Biotech

Commercial FTO Checklist

Freedom to operate when using generative AI in drug discovery requires a structured IP risk assessment that extends beyond traditional compound FTO searches. The following ten‑point checklist is designed for biotech R&D teams and their in‑house counsel:

1. Training data rights. Confirm that all datasets used to train the AI model are properly licensed and that licence terms permit commercial use of model outputs.
2. Model licence terms. Review the licence for any third‑party AI model or foundation model to determine whether generated outputs are freely assignable or subject to restrictions.
3. Third‑party patent landscape. Conduct a targeted patent search covering AI methods applied in the relevant therapeutic area, focusing on method claims that could read on your pipeline.
4. Output novelty check. Search the prior art for the specific compound, sequence or composition generated by the AI to confirm novelty before committing to development.
5. Inventorship documentation. Maintain contemporaneous records of human contributions to the inventive process, selection criteria, validation decisions, experimental design, to support inventorship declarations.
6. Model provenance and versioning. Track model versions, hyperparameters and training runs used to generate each candidate, ensuring reproducibility and supporting patent disclosure requirements.
7. Regulatory data requirements. Assess whether regulatory agencies (MHRA, EMA, FDA) require disclosure of AI involvement in candidate selection and ensure compliance with any emerging obligations.
8. Open‑source component audit. Identify any open‑source software incorporated into the AI pipeline and verify that licence terms (e.g., GPL, Apache 2.0) are compatible with commercial patent filings.
9. Contractual IP allocation. In collaborations and CRO arrangements, ensure that contracts clearly allocate IP rights in AI‑generated outputs and address jointly generated inventions.
10. Ongoing monitoring. Establish a watching brief for newly published patent applications in the AI‑drug‑discovery space that could create future FTO risks.

IP Risk Matrix

Practical Workflow for a Biotech R&D Team

An effective workflow begins at the earliest stage of AI‑assisted research. When the R&D team initiates a new AI‑driven discovery campaign, IP counsel should be embedded in the project planning process to: (1) define the scope of human inventive contribution expected at each stage; (2) set documentation standards for model inputs, outputs and validation experiments; and (3) trigger a preliminary patent landscape search before resources are committed to lead optimisation. This integrated approach reduces the risk of discovering FTO obstacles, or enablement gaps, only after significant R&D investment has been made.

7. Conclusion: Immediate Actions for Counsel Advising on AI Patentability UK

The 2026 UK Supreme Court judgment and subsequent UKIPO guideline updates represent the most significant shift in AI patentability UK practice in over a decade. For life‑science companies, the practical landscape is clearer: AI‑assisted inventions are patentable if they meet standard patentability criteria, provided the claims are anchored in a demonstrable technical contribution and supported by enabling disclosure. The alignment with EPO practice simplifies dual‑jurisdiction prosecution and reduces the need for UK‑specific claim workarounds.

Three immediate steps for in‑house counsel and patent attorneys:

• Audit existing portfolios and pending applications. Identify any UK patent applications that were refused or narrowed under the old signpost test and assess whether refiled or divisional applications could benefit from the new framework.
• Update drafting protocols. Ensure all new patent specifications for AI‑assisted inventions include sufficient model disclosure, experimental validation data and claims that tie the AI component to a technical effect in the biological context.
• Conduct a freedom‑to‑operate review. With the bar to obtaining AI patents now lower, the competitive landscape will expand. Commission targeted FTO searches for AI method patents in your therapeutic areas and establish a watching brief for newly published applications.

For tailored advice on patent prosecution strategy, portfolio review or freedom‑to‑operate analysis in the AI and life‑sciences space, explore the United Kingdom lawyer directory to connect with experienced patent practitioners.

Sources

1. GOV.UK, Guidelines for Examining Patent Applications Relating to AI Inventions (UKIPO)
2. Osborne Clarke, Supreme Court Opens the Door in the UK to Patentability of AI Inventions
3. Mishcon de Reya, UK Supreme Court Reshapes the Approach to Patentability of AI Inventions and Beyond
4. DLA Piper, Patentability of AI Inventions in the UK, US and China
5. Hepworth Browne, Can AI Inventions Be Patented in the UK? A Complete Post‑UKSC 3 Guide
6. Morgan Lewis, UK Supreme Court Resets Approach to Patentability of Computer‑Implemented Inventions
7. EPO Guidelines for Examination
8. WP Thompson, The UK Just Made It Easier to Patent AI Inventions